Qvikhr 3 1.7B Instruction Noreasoning

🚀 QVikhr-3-1.7B-Instruction-noreasoning

This is an instructive model based on Qwen/Qwen3-1.7B, trained on the Russian-language dataset GrandMaster2. It is designed for high-efficiency text processing in Russian and English, delivering precise responses and fast task execution.

🚀 Quick Start

The model is ready to use for high - efficiency text processing in Russian and English. You can refer to the sample code below to start using it.

✨ Features

• Base Model: Qwen/Qwen3-1.7B
• Specialization: RU
• Dataset: GrandMaster2
• Support: Bilingual RU/EN

📦 Quantized variants

• GGUF Vikhrmodels/QVikhr-3-1.7B-Instruction-noreasoning-GGUF
• MLX
  • 4 bit Vikhrmodels/QVikhr-3-1.7B-Instruction-noreasoning-MLX_4bit
  • 8 bit Vikhrmodels/QVikhr-3-1.7B-Instruction-noreasoning-MLX_8bit

💻 Usage Examples

Basic Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

# Load the model and tokenizer
model_name = "Vikhrmodels/QVikhr-3-1.7B-Instruction-noreasoning"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Prepare the input text
input_text = "–ù–∞–ø–∏—à–∏ –∫—Ä–∞—Ç–∫–æ–µ –æ–ø–∏—Å–∞–Ω–∏–µ –∫–Ω–∏–≥–∏ –ì–∞—Ä—Ä–∏ –ü–æ—Ç—Ç–µ—Ä."

messages = [
    {"role": "user", "content": input_text},
]

# Tokenize and generate text
input_ids = tokenizer.apply_chat_template(messages, truncation=True, add_generation_prompt=True, return_tensors="pt")
output = model.generate(
    input_ids,
    max_length=1512,
    temperature=0.3,
    num_return_sequences=1,
    no_repeat_ngram_size=2,
    top_k=50,
    top_p=0.95,
)

# Decode and print result
generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
print(generated_text)

Model response

<think>

</think>

The response in the original text is in an unreadable format, and it seems to be in a non - standard encoding. Please check and provide the correct text if you need a more accurate presentation.

📚 Documentation

Description

QVikhr-3-1.7B-Instruction-noreasoning is a robust language model trained on the GrandMaster-2 dataset. It excels in instruction generation, contextual responses, and text analysis in Russian. The model is optimized for instructional tasks and textual data processing, suitable for professional use as well as integration into user - facing applications and services.

Training

QVikhr-3-1.7B-Instruction-noreasoning was developed using the SFT (Supervised Fine - Tuning) FFT (Full Fine - Tune) method. The synthetic dataset GrandMaster-2 was used for training.

📊 Ru Arena General

📄 License

The model is released under the apache-2.0 license.

👨‍💻 Authors

• Sergei Bratchikov, NLP Wanderer, Vikhr Team
• Nikolay Kompanets, LakoMoor, Vikhr Team
• Konstantin Korolev, Vikhr Team
• Aleksandr Nikolich, Vikhr Team

📖 How to Cite

@inproceedings{nikolich2024vikhr,
  title={Vikhr: Advancing Open-Source Bilingual Instruction-Following Large Language Models for Russian and English},
  author={Aleksandr Nikolich and Konstantin Korolev and Sergei Bratchikov and Nikolay Kompanets and Igor Kiselev and Artem Shelmanov},
  booktitle={Proceedings of the 4th Workshop on Multilingual Representation Learning (MRL) @ EMNLP-2024},
  year={2024},
  publisher={Association for Computational Linguistics},
  url={[https://arxiv.org/pdf/2405.13929](https://arxiv.org/pdf/2405.13929)}
}

@misc{qwen3technicalreport,
      title={Qwen3 Technical Report}, 
      author={Qwen Team},
      year={2025},
      eprint={2505.09388},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.09388}, 
}

⚠️ Important Note

The recommended generation temperature is 0.3.